User plane and control plane separation based on mobility and throughput prediction

This application is a 35 U.S.C. § 371 national stage application of PCT International Application No. PCT/EP2021/069621 filed on Jul. 14, 2021, the disclosure and content of which is incorporated by reference herein in its entirety.

The present invention relates to methods for controlling wireless communication and to corresponding devices, systems, and computer programs.

rd In wireless communication networks, e.g., as specified by 3GPP (3Generation Partnership Project), data traffic may be differentiated into user plane (UP) traffic and control plane (CP) traffic. The UP traffic includes user data generated by services or applications executed on a user device, typically referred to as UE (“user equipment”). The CP traffic includes control signaling, such as data for authentication, mobility management, billing, or the like.

3GPP TS 23.214 V16.2.0 (2020-09) specifies an architecture enhancement denoted as CP-UP separation (CUPS), which aims at enabling network operators to flexibly deploy and operate their networks to serve the needs of specific applications and reduce cost. For example, for low-latency applications, e.g., such as ultra-reliable low latency communication (uRLLC) applications, UP nodes or UP network functions can be placed closer to the radio edge, while CP nodes or CP network functions can be placed in a more centralized way. In addition to positioning, CUPS allows for independent scaling of the nodes, for example for use cases that require more UP traffic than CP traffic, such as some enhanced mobile broadband (eMBB) applications, it is possible to scale UP functions independently of CP functions. On the other hand, for applications that require more CP traffic than UP traffic, such as in the case of machine-to-machine (M2M) applications, it is possible to scale CP functions while leaving UP functions intact.

In the LTE (Long Term Evolution) technology specified by 3GPP, dual connectivity of a UE enables the UE to connect to two different access nodes, in the LTE technology denoted as eNB (“enhanced Node B”). This may in turn allow the UE to enhance its data throughput by using carrier aggregation with respect to the multiple eNBs. Further, multi-connectivity may allow a UE to simultaneously connect to multiple access nodes of different radio access technologies (RATs). This may for example allow for achieving redundancy or load splitting. Further, dual-connectivity or multi-connectivity also supports that CP traffic of a UE is transmitted with respect to one access node while UP traffic is transmitted with respect to another access node.

In “Energy efficiency optimization-oriented control plane and user plane adaptation with a frameless network architecture for 5G” by Xu, X., Dai, X., Liu, Y. et al., Journal on Wireless Communication Networking 2015, 159 (2015), https://doi.org/10.1186/s13638-015-0403-5, an end-to-end CUPS algorithm is proposed which reduces transmission power for RBSs (Radio Base Stations). In “Performance Analysis of Cellular Radio Access Networks Relying on Control- and User-Plane Separation” by K. Liang, G. Liu, L. Zhao, X. Chu, S. Wang and L. Hanzo, IEEE Transactions on Vehicular Technology, vol. 68, no. 7, pp. 7241-7245 July 2019, a CUPS algorithm is proposed that aims at providing reliable coverage for CP traffic and highest energy efficiency for UP traffic. This algorithm is based on the assumption that CP traffic is comparatively low throughput as compared to UP traffic.

However, these known CUPS algorithm may still have deficiencies in optimizing energy efficiency for both CP traffic and UP traffic. Accordingly, there is a need for techniques which allow for improving energy efficiency in CUPS scenarios.

According to an embodiment, a method of controlling wireless communication is provided. The method comprises, for a set of access nodes of a wireless communication network, predicting data throughput of UP traffic and CP traffic generated by wireless communication devices being served by the access nodes on a plurality of wireless carrier signals. Further, the method comprises predicting mobility of the wireless communication devices. Further, the method comprises, based on the predicted data throughput and the predicted mobility, deciding for at least some of the plurality of wireless carrier signals whether the wireless carrier signal is assigned either to serving UP traffic, to serving CP traffic, or to serving both UP traffic and CP traffic.

According to a further embodiment, a node for a wireless communication network is provided. The node is configured to, for a set of access nodes of a wireless communication network, predict data throughput of UP traffic and CP traffic generated by wireless communication devices being served by the access nodes on a plurality of wireless carrier signals. Further, the node is configured to predict mobility of the wireless communication devices. Further, the node is configured to, based on the predicted data throughput and the predicted mobility, decide for at least some of the plurality of wireless carrier signals whether the wireless carrier signal is assigned either to serving UP traffic, to serving CP traffic, or to serving both UP traffic and CP traffic.

According to a further embodiment, a node for a wireless communication network is provided. The node comprises at least one processor and a memory. The memory contains instructions executable by said at least one processor, whereby the node is operative to, for a set of access nodes of a wireless communication network, predict data throughput of UP traffic and CP traffic generated by wireless communication devices being served by the access nodes on a plurality of wireless carrier signals. Further, the memory contains instructions executable by said at least one processor, whereby the node is operative to predict mobility of the wireless communication devices. Further, the memory contains instructions executable by said at least one processor, whereby the node is operative to, based on the predicted data throughput and the predicted mobility, decide for at least some of the plurality of wireless carrier signals whether the wireless carrier signal is assigned either to serving UP traffic, to serving CP traffic, or to serving both UP traffic and CP traffic.

According to a further embodiment of the invention, a computer program or computer program product is provided, e.g., in the form of a non-transitory storage medium, which comprises program code to be executed by at least one processor of one or more nodes of a wireless communication network. Execution of the program code causes the one or more nodes to, for a set of access nodes of a wireless communication network, predict data throughput of UP traffic and CP traffic generated by wireless communication devices being served by the access nodes on a plurality of wireless carrier signals. Further, execution of the program code causes the one or more nodes to predict mobility of the wireless communication devices. Further, execution of the program code causes the one or more nodes to, based on the predicted data throughput and the predicted mobility, decide for at least some of the plurality of wireless carrier signals whether the wireless carrier signal is assigned either to serving UP traffic, to serving CP traffic, or to serving both UP traffic and CP traffic.

Details of such embodiments and further embodiments will be apparent from the following detailed description of embodiments.

In the following, concepts in accordance with exemplary embodiments of the invention will be explained in more detail and with reference to the accompanying drawings. The illustrated embodiments relate to management of DL transmit powers in a wireless communication network. The wireless communication network may be based on various technologies, e.g., the NR (New Radio) technology specified, the LTE technology specified by 3GPP, the UMTS (Universal Mobile Telecommunications System) technology specified by 3GPP, the General Packet Radio System (GPRS) technology specified by 3GPP, or on a combination of two or more of these technologies.

The illustrated concepts aim at optimization of energy efficiency of CUPS in a set of access nodes operating on a set of wireless carrier signals, in the following also denoted as carriers. The access nodes are herein also denoted Here, it is noted that different access nodes may utilize different carriers, and that one or more of carriers may be assigned to each access node. Further, each carrier may be assigned to a certain traffic direction, e.g., to a downlink (DL) direction from the wireless communication network to the UE or to an uplink (UL) traffic direction from the UE to the wireless communication network. Some of the carriers may also be assigned to both the UL traffic direction and the DL traffic direction. In the following such carriers are also denoted as UL carriers, DL carriers, and UL/DL carriers, respectively. The access nodes may support different RATs, e.g., the NR RAT, the LTE RAT, the UMTS RAT, and/or the GPRS RAT.

For the optimization of energy efficiency of CUPS, a CUPS optimization algorithm predicts data throughput of UP traffic and CP traffic generated wireless communication devices, in the following denoted as UEs, which are served by the set of access nodes. Further, the CUPS optimization algorithm predicts mobility of the UEs, e.g., attachment, detachment, or handovers, with respect to the access nodes of the set. Based on the predicted data throughput and the predicted mobility, the CUPS optimization algorithm decides for at least some of the carriers whether the carrier is assigned to UP traffic, to CP traffic, or to both UP traffic and CP traffic. Such assignment is in the following also denoted as “role assignment”. The role assignment decisions may be performed with the aim of minimizing overall energy consumption of the access nodes, however while maintaining coverage of the access nodes, i.e., without losing coverage in a nominal coverage area of the access nodes. The CUPS optimization algorithm may be regarded as having two stages: a prediction stage which provides predictions of data throughput and mobility, and an assignment optimization stage, which optimizes the role assignments of the carriers and may also decide whether certain ports associated with the carriers can be turned off or need to be activated.

1 FIG. 1 FIG. 1 FIG. 1 FIG. 10 100 100 100 100 210 210 220 230 220 10 10 10 10 10 230 230 230 illustrates exemplary structures of the wireless communication network. In particular,shows multiple UEswhich are served by access nodesof the wireless communication network. Here, it is noted that each access nodemay serve a number of cells within the coverage area of the wireless communication network. The depending on the supported RAT(s) each access nodesmay correspond to a gNB (“next generation Node B”) of the NR technology, an eNB of the LTE technology, an NB (“Node B”) of the UMTS technology, and/or a RBS (“Radio Base Station”) of the GPRS technology. The access nodesmay each be regarded as being part of an RAN (Radio Access Network) of the wireless communication network. Further,schematically illustrates a CN (Core Network)of the wireless communication network. In, the CNis illustrated as including a GW (gateway), and a controller (CTRL). The GWmay be responsible for handling UP traffic of the UEs, e.g., by forwarding user data traffic from a UEto a network destination or by forwarding user data traffic from a network source to a UE. Here, the network destination may correspond to another UE, to an internal node of the wireless communication network, or to an external node which is connected to the wireless communication network. Similarly, the network source may correspond to another UE, to an internal node of the wireless communication network, or to an external node which is connected to the wireless communication network. The controllermay in turn be responsible for controlling the user data traffic, e.g., with respect to connection establishment, authentication, and/or mobility. The controllermay thus be responsible for handling CP traffic of the UEs. The controllercould for example correspond to an AMF (Access and Mobility Management Function) of the NR technology, an MME (Mobility Management Entity) of the LTE technology.

100 100 210 250 210 210 260 210 260 210 250 260 260 120 10 100 10 260 10 10 10 100 270 210 270 210 100 1 FIG. 1 FIG. 1 FIG. As illustrated by double-headed arrows, the access nodesmay send DL transmissions to the UEs, and the UEs may send UL transmissions to the access nodes. The DL transmissions and UL transmissions may be used to provide various kinds of services to the UEs, e.g., a voice service, a multimedia service, or a data service. Such services may be hosted in the CN, e.g., by a corresponding network node. By way of example,illustrates an application service platformprovided in the CN. Further, such services may be hosted externally, e.g., by an AF (application function) connected to the CN. By way of example,illustrates one or more application servers platformprovided connected to the CN. The application server(s)could for example connect through the Internet or some other wide area communication network to the CN. The application service platformmay be based on a server or a cloud computing system and be hosted by one or more host computers. Similarly, the application serversmay be based on a server or a cloud computing system and be hosted by one or more host computers. The application server(s)may include or be associated with one or more AFs that enable interaction with the CNto provide one or more services to the UEs, corresponding to one or more applications. These services or applications may generate the user data traffic conveyed by the DL transmissions and/or the UL transmissions between the access node(s)and the respective UE. Accordingly, the application servermay include or correspond to the above-mentioned network destination and/or network source for the user data traffic. In the respective UE, such service may be based on an application (or shortly “app”) which is executed on the UE. Such application may be pre-installed or installed by the user. Such application may generate at least a part of the UP traffic between the UEand the access node. Further,illustrates a management systemprovided in the CN. The management systemmay be responsible for various management functionalities, such as configuration of nodes of the CNor of the RAN, in particular configuration of the access nodes.

2 2 FIGS.A andB 2 2 FIGS.A andB 1 FIG. 2 FIG.A 2 FIG.B 1 FIG. 100 10 100 200 100 100 200 100 200 100 100 200 200 100 200 200 200 100 100 100 200 210 210 270 schematically illustrates functionalities of the CUPS optimization algorithm. In particular,illustrate a set of access nodesto be optimized, UEsserved by these access nodes, and a logical component denoted as role assignment function (RAF), which implements the CUPS optimization algorithm. The access nodescould for example correspond to any of the access nodesillustrated in. The RAFis aware of the configuration of the access nodesof the set. Further, the RAFretrieves data traffic information from the access nodesas well as an energy consumption profile of each access node. Using the reported data traffic information, the RAFpredicts future data traffic. Based on the predicted future data traffic, the RAFthen decides whether the respective role of one or more of the access nodesshould be reassigned, taking into account the energy profiles of the access nodes.illustrates a variant where the RAFis assumed to be part of the RAN. For example, the RAFcould be co-located with one of the access nodes, be implemented by one of the access nodes, or be implemented as a cloud system distributed over two or more of the access nodes.illustrates a variant where the RAFis provided outside the RAN, e.g., in the CNor even in a third-party node outside the CN, e.g., in the management systemof.

3 FIG. 100 For the sake of better explaining the illustrated concepts,schematically illustrates flows of data signals and power in an access node (AN). The AN may correspond to one of the above access nodes. Power flows are illustrated by solid arrows and data signal flows are illustrated by broken lines.

3 FIG. 310 310 1 310 2 310 320 320 1 320 2 330 340 350 320 310 310 320 10 340 210 In the example of, the AN is assumed to be provided with multiple antennas, in particular a first antennadenoted as Antennaand a second antennadenoted as Antenna. For each of the antennas, the AN includes a corresponding radio unit, in particular a radio unitdenoted as Radio Unitand a second radio unitdenoted as Radio Unit. Further, the AN includes baseband circuitry, a switch, and a power subsystem. It is noted that these components do not necessarily constitute a unitary device. For example, the radio unitsand the antennascould be located remotely from the other components, e.g., on a radio tower. The antennasand the radio unitsprovide a radio interface (Radio IF) of the AN, which is used for connecting to one or more UEs, such as the above UEs. The switchprovides a backhaul interface (Backhaul IF) of the AN, which is used for connecting the AN to other nodes of the wireless communication network, e.g., nodes of the CNor other ANs.

350 250 340 330 320 310 320 320 330 340 210 310 3 FIG. The power subsystemis assumed to be powered by mains power from the grid and/or from renewable sources. In some cases, powering of the power subsystem can also be assisted by batteries. The power subsystemdirectly or indirectly powers the switch, baseband circuitry, the radio units, and the antennas, and optionally also other components which are not illustrated in, e.g., cooling equipment. In the DL traffic direction, the radio unitconverts a radio data signal received from the corresponding antennato an electrical data signal and provides it for further processing to the baseband circuitry. The baseband circuitry then sends the processed data signal in digital form to the switch, typically to be forwarded to the CNand then to its final destination. In the UL traffic direction, the data follows the opposite path. For conveying the electrical data signals, each radio is connected via multiple RF (radio frequency) ports to its corresponding antenna. This connection is typically provided by coaxial cables.

3 FIG. 320 320 320 320 As illustrated in, each radio unitcan have more than one pair of RF ports. Such multiple pairs of RF ports can be used to support more than one RAT in the AN, e.g., GPRS and LTE, or LTE and NR. Each radio unitmay also have a certain power efficiency rating, for example expressed in mW/Kbit. In addition to the power efficiency rating, the power efficiency of the radio unitalso depends on the current radio load. Accordingly, power efficiency of the radio unitcan be described as in terms of the power efficiency rating per radio load. The radio load can be described as percentage of a maximum load, e.g., in terms of active UEs as percentage of maximum UEs that can be supported, or in terms of a percentage of used bandwidth. In typical ANs, for example up to 40 W can be allocated per RF port of a radio unit with four RF ports, i.e., 4 times 40 W.

200 100 200 In the following, the illustrated concepts will be further explained by referring to a more detailed exemplary implementation of the CUPS optimization algorithm. As mentioned above, the RAFmay be provided with information concerning the configuration of the access nodes. This information may specifically include data about sectors served by the AN, such as the amount of sectors and the amount of radio units (RUs) in each sector. Since every RU typically also has a number of RF ports, each port serving a specific RAT on a number of carriers, the information provided to the RAFmay also indicate the number of RF ports of each RU and the served RAT. Still further, the information provided to the RAF may indicate the number of UEs attached on a certain carrier thus generating data traffic on the carrier. As mentioned above, each carrier may be configured to serve CP traffic, UP traffic, or both CP traffic and UP traffic.

The RU has the capacity to turn off unused RF ports independently, thereby allowing to save energy. The power efficiency rating of each RU typically relates to a number of factors such as the model of the RU, associated circuitry of the AN, or utilized type of power supply unit (PSU) or power distribution unit (PDU). Further, the power efficiency of the RU depends on the actual data throughput of the RU, i.e., the load pf the RU, environmental conditions, such as temperature and humidity, and the stability of the mains power, which may for example depend on the type of power source, such as a combination of batteries and own renewables, or power grid.

POP 1 k x POP 200 x x Some static information such as the ambient temperature (AN_temp) and humidity (AN_hum). x x Y x k x x k x k x 1 A collection of RUs RUPOP_AN={RU_AN, . . . , RU_AN}, and for each RU_ANin RUPOP_ANa number of RF ports, each port having a state, such as active, idle, or shut, a number of carriers, and for every carrier an indication of its RAT, e.g., UMTS, LTE, or NR, the current data throughput on the carrier, differentiated as UL CP throughput, DL CP throughput, UL UP throughput, and DL UP throughput, a list of active UEs on the carrier, and the energy efficiency rating of RU_AN, e.g., in terms of watts/kbit. A corresponding information element have may the form: RU_AN={Power_Source, port1[state, carrier1{RAT, throughput[CP[UL][DL]][UP[UL][DL]], UE_list, Power_Efficiency}, . . . carrierL{RAT, throughput[CP[UL][DL][UP[UL][DL], UE_list, Power_Efficiency}], port2[ . . . ], . . . , portj[ . . . ]}. ID 210 A list of active UEs on each carrier, corresponding to the above UE_list. Each UE in the list may be identified by a UE identifier (UE), e.g., its IMSI (International Mobile Subscription Identifier). For each UE, the list includes the UE's recent mobility history. The recent mobility history can be expressed as a sequence of mobility events including handovers from one cell to another, attachment to a cell, or detachment from a cell. These cells may each be identified by a cell identifier (cellID). Information about the location of the cell corresponding to a certain cellID is typically stored in the mobility management nodes of the CN, i.e., in an MME of the LTE technology or an AMF of the NR technology. Accordingly, the mobility history contained in the list of active UEs can be a list of 4-tuples having the form: <cellID, latitude, longitude, timestamp>. For a more mathematical description, the set of ANs to be optimized can be expressed as a population of neighboring ANs: AN={AN, . . . , AN}. Then, for every AN∈ANthere the following information may be reported to RAF:

4 FIG. 400 400 410 420 410 420 predictor predictor shows a block diagram for illustrating the CUPS optimization algorithm, in the illustration denoted by. As illustrated, the CUPS optimization algorithmutilizes a first machine learning (ML) modelfor the prediction of data throughputs, in the following also denoted as trafficand a second ML modelfor prediction of UE mobility, in the following also denoted as mobility. In the illustrated example, the first ML modeland the second ML modelare each assumed to be based on a neural network. However, it is noted that other types of ML model or artificial intelligence (AI) based mechanism could be used as well.

410 410 400 410 410 410 As illustrated, the first ML modeloperates on the basis of throughput data representing data throughputs observed for each UE on the list of active UEs. The throughput data are differentiated into UL CP throughput, DL CP throughput, UL UP throughput, and DL UP throughput. The first ML modelis also trained on the basis of such throughput data, e.g., in a dedicated training phase and/or during regular operation of the CUPS optimization algorithm. For training, newly received throughput data may be used for verifying the predictions made by the first ML modeland to provide feedback for correcting neural network weights of the first ML model. Based on its input, the first ML modelprovides a throughput prediction representing future throughputs of each UE on the list of active UEs, e.g., for the next 24 hours.

420 420 400 420 420 420 The second ML modeloperates on the basis of mobility data representing mobility observed for the UEs on the list of active UEs. The second ML modelis also trained on the basis of such mobility data, e.g., in a dedicated training phase and/or during regular operation of the CUPS optimization algorithm. For training, newly received mobility data may be used for verifying the predictions made by the second ML modeland to provide feedback for correcting neural network weights of the second ML model. Based on its input, the second ML modelprovides a mobility prediction representing future mobility of each UE on the list of active UEs, e.g., for the next 24 hours. Such mobility may be due to handovers, which typically result in the UE moving from one carrier to another carrier. Further, the mobility involve attachment of a UE and/or detachment of a UE. Each of such mobility events thus has an effect on the UEs which are active on a certain carrier.

400 430 410 420 430 430 430 As further illustrated, the CUPS optimization algorithmincludes a CP-UP assignment optimization stagewhich receives the throughput predictions from the first ML modeland the mobility predictions from the second ML model. Based on these predictions, the CP-UP assignment optimization stageadapts the role assignments of the carriers, i.e., assigns one or more of the carriers to UP traffic, assigns one or more of the carriers to CP traffic, or assigns one or more of the carriers to both UP traffic and CP traffic. Such assigned carriers are herein also denoted as CP carrier, UP carrier, and CP/UP carrier, respectively. The adaptation of the CP-UP assignment aims at reducing the overall energy consumption of the ANs. At its output, the CP-UP assignment optimization stagethus provides CP-UP assignments indicating the assignment of each carrier. These CP-UP assignments may then be used for correspondingly reconfiguring the ANs. Alternatively or in addition, these CP-UP assignments may be used as input or a further iteration of the CP-UP assignment optimization stage. Based on the adapted role assignments, it can be decided to activate or deactivate RF ports of the ANs, to initiate handovers between carriers, within the same AN or different ANs, and/or to notify UEs about the reconfiguration of the carriers.

430 It is noted that also the CP-UP assignment optimization stagecould be based on an ML model. Such ML model could be trained using reinforcement learning (RL), based on actions corresponding to role assignments of the carriers and rewards based on changes in energy consumption reported by the considered ANs.

400 In an illustrative example, the operation of the CUPS optimization algorithmmay be described by the following pseudo-code:

predictor Initialize trafficneural network weights or alternatively use weights from the previous iteration predictor Initialize mobilityneural network weights or alternatively use weights from previous iteration From time T=0 to T=K POP  Collect mobility information for each UE from each AN in ANand create mobility profile of ID UE: [list (UE, list [latitude, longitude, time]]  Collect throughput information for UE from every RBS in RBSPOP and create traffic profile: ID [list (UE, list [ul_throughput, downlink_throughput, time]] predictor predictor  Train the trafficand mobilityusing collected mobility profile and traffic profile data After time K elapses: POP  Retrieve ANdata, e.g., via X2 protocol from neighboring ANs or via OSS (Operations Support System) Initialize predictionsLIST to empty current POP For every ANin AN current current  For every RUin AN current current   For every portin RU current current    For every carrierin port prediction     Initialize throughput[CP[UL][DL], UP[UL][DL]] prediction     Initialize            mobility [[N,wN][NE,wNE][E,wE][SE,wSE][S,wS][SW,wSW][W,wW][NW,wNW]] predictor current prediction     Execute trafficwith carrier.throughput and update throughput current input     For every UE in carrier, iterate UE_list and create a mobility= N NE E SE S SW W NW No — Change [[N,w][NE,w][E,w][SE,w][S,w][SW,w][W,w][NW,w][No_change, w]] predictor input prediction     Execute mobilitywith mobility, and update mobility current ID current ID current prediction     Add <ANAN, RURU, carriercarrierID, mobility, prediction throughput> to predictionsList Iterate predictionsList and extract predicted throughput per RU and per AN, differentiated by prediction CP, UP, UL, and DL, i.e. throughput[CP[UL][DL], UP[UL][DL]] and identify expected prediction demand using predicted mobility patterns mobility

Reassign RU carrier roles as CP carrier, UP carrier or CP/UP carrier for UL and DL traffic direction based on the predicted throughput demand taking into account the predicted mobility patterns, starting from the RUs with best power efficiency and most stable power source.

400 400 410 420 410 420 400 410 420 The CUPS optimization algorithmmay be iterated every time K, e.g., every day. The CUPS optimization algorithmmay be executed in in two phases: In a first phase, the mobility data and throughput data is collected from the ANs in the population. These data are buffered, and when the time K elapses, the data is used to train the first ML modeland the second ML model. The first ML modeland the second ML modelcan be randomized in each iteration of the CUPS optimization algorithm. Alternatively, the first ML modeland the second ML modelcan be trained incrementally using a transfer learning approach, i.e., learned results from previous iterations can be maintained. The choice between randomization and incremental learning may depend on dynamics of the data traffic and mobility. In very dynamic environments, e.g., cities, randomization and learning from scratch may be preferable in order not to carry over bias from previous training. In less dynamic environments, e.g., rural areas or for applications that have fixed UE traffic and mobility profiles such as certain M2M applications with sensors, incremental learning may be more preferable. In each case, the learning may be based on data reported and collected per UE:

410 For the first ML modelfor the prediction of throughput, data elements of the collected data consist of the UE identifier, the data throughput generated by the UE in the UL traffic direction and the DL traffic direction, differentiated by CP traffic and UP traffic, and a timestamp indicating when the data throughput was observed. Such information may for example be available from counters in the ANs.

420 N NE E SE S SW No_Change For the second ML modelfor the mobility prediction, data elements of the collected data consist of the UE identifier, a cell identifier of the cell that the UE is attached to, such as a cell global identity, as well as a timestamp indicating the time when these data were observed. Such data may be available from mobility management nodes, such as an MME or AMF, or via OSS. Further, such data may be translated into geographical position information, in particular latitude and longitude. From changes of such data mobility events like handovers, attachment of the UE, or detachment of the UE may be derived. Further, changes of the geographical positions may be used to derive mobility indicators like movement toward north ([N]) and velocity of movement toward north ([w]), movement toward northeast ([NE]) and velocity of movement toward northeast ([w]), movement toward east ([E]) and velocity of movement toward east ([w]), movement toward southeast ([SE]) and velocity of movement toward southeast ([w]), movement toward south ([S]) and velocity of movement toward south ([w]), movement toward southwest ([SW]) and velocity of movement toward southwest ([w]), and no change of position ([No_Change]) and velocity of no change of position ([w]). Here, it is noted that even when not changing the position, i.e., in the mobility state “no change of position”, the UE may move with a certain velocity, however without moving to another cell.

410 420 The output of the first ML modeland the second ML modelis a predicted data throughput of UEs and a predicted mobility pattern of the UEs. The output can be based on statistic evaluation by regression or on classification. Here, usage of classification may in some cases allow for less complexity.

5 FIG. 5 FIG. 410 420 illustrates a table with an example of input data and output data of the first ML modeland the second ML model. In the example of, it can be seen from the throughput data that for one of the data elements, the UE with UE identity 082902022557309891 is expected to generate fairly low throughput of CP traffic and rather high throughput of UP traffic. From the mobility data collected at the same time, it is evident that the UE with UE identity 082945447387771631 is likely to move in the northern direction.

410 420 410 420 410 420 410 420 430 410 420 The neural networks used in the first ML modeland the second ML modelshould be sensitive to time series datasets. Accordingly, recurrent neural networks (RNNs) such as long-short term memory networks (LSTMs) may be used in the first ML modeland the second ML model. The training of the first ML modeland the second ML modelmay be conducted by splitting the dataset received in the buffer in a random way to a set of training data and a set of test data used for verification and testing of the trained model's performance using an accuracy metric, e.g., area under curve (AUC). The training process may conclude if the metric indicates satisfactory accuracy of prediction. Accordingly, the first and second ML models,, which are responsible for predicting throughput and mobility, may be based on a different type of ML model than the CP-UP assignment optimization state. In particular, the first ML modeland the second ML modelmay be based on RNNs, while the CP-UP optimization stage is based on RL.

400 400 410 420 an estimate of future total bandwidth on a carrier for CP traffic and UP traffic in the UL traffic direction and the DL traffic direction, e.g., more than 10 Mbit for UP traffic in the DL traffic direction and less than 2 Mbit for UP traffic in the UL direction, and an estimate of future mobility of UEs on the carrier, e.g., indicating that 90% of the UEs are expected to move North, 5% of the UEs are expected to move south, and 5% are expected to move east. The next phase of the CUPS optimization algorithmis a deployment phase, in which CUPS optimization algorithmdecides on the role assignments of the carriers, using the predictions from the first ML modeland the second ML modelHere, it is noted that in the illustrated example the CP-UP role assignment is done wireless carrier for each RF port of each RU of each AN in the considered population of ANs. The process follows an approach wherein current observed values at time K, e.g., observations of last 10 minutes for K=1 day, for mobility and throughput are used to predict future mobility and throughput per UE. Then, the throughput and mobility predictions are aggregated for all UEs in a carrier. At the end of this aggregation step, the following information is produced per carrier:

400 400 400 400 400 Based on such predictions, the CUPS optimization algorithmconducts a two-step CP-UP reassignment procedure per carrier. First, the CUPS optimization algorithmidentifies which ANs are more likely to have a high throughput demand by looking at the aggregate estimates of predicted mobility. For example if UEs in the middle of the coverage area of the AN population are moving north, then the ANs in the northern part of the coverage area can be identified being likely to have a high throughput demand. Then the CUPS optimization algorithmmay further analyze the expected throughput demand by predicting bandwidth will be needed for CP traffic in the DL traffic direction, CP traffic in the UL traffic direction, UP traffic in the DL traffic direction, and UP traffic in the UL traffic direction. Then the CUPS optimization algorithmidentifies which carriers are more energy efficient and reassigns the roles of the carriers if necessary. This reassignment process may start with reassignment of the of the most energy efficient carrier by assigning this carrier to handle the largest volume of CP traffic or UP traffic according to the predicted throughput demand. When performing the reassignment of the roles, the CUPS optimization algorithmensures that no part of the overall coverage areas of the considered population of ANs is left without coverage for CP traffic or UP traffic in any traffic direction. This can for example be achieved by reserving at least one carrier for that type of traffic which has a risk of losing coverage when considering merely the energy efficiency. After the reassignment, some RF ports of the ANs may be turned off if they are not needed for serving the throughput demand or maintaining the coverage.

400 On the side of the UEs, the CUPS optimization algorithmmay benefit from the UEs being able to connect to multiple RATs, e.g., using multiple radios. In this case, the reassignment of the carrier roles may consider that for a certain traffic type, e.g., UP traffic in the DL direction, a carrier of another RAT may provide a better energy efficiency. It is however noted that improvement of energy efficiency is also possible within the same RAT, e.g., due to different types of RUs which may have different energy efficiency rating for the same RAT.

400 In the following, some examples will explained in order to illustrate the energy benefits which can be obtained CUPS optimization algorithm. In these examples, a population of three ANs is assumed. Each AN is assumed to have three sectors, each covering 120 degrees, and RU per sector. Each RU has four RF ports, and a maximum of 40 W can be allocated to each RF port. Further, it is assumed that in each RU two RF ports, one transmit (TX) port and one reception (RX) port (RX/TX) are allocated to the GPRS RAT and that further two RF ports, one TX port and one RX port, are allocated to the LTE RAT. The cost for operating an idle RF port is assumed to be 10 W per RF port, to which the cost of data traffic is added up to the maximum of 40 W.

24 The GPRS carriers are assumed to be in the 900 MHz band, e.g., 890-915 MHz for TX carriers and 935-960 MHz for RX carriers. The LTE carriers are assumed to be in the band, i.e., 1626.5-1660.5 MHz for TX carriers and 1525-1559 MHz for RX carriers.

The LTE RAT typically has a range of up to 2 km, while the GPRS RAT has a range of 10 km. In the illustrated example, it is further assumed that the power efficiency of an LTE carrier is 10 mW/kbps, whereas the power efficiency of a GPRS is 2 mW/kbps. For reasons of simplicity it is also assumed that all carriers of the same RAT have the same power efficiency and that the power efficiency is independent of the load. For the GPRS RAT a capacity of 25 kbps per UE is assumed, corresponding to a baseline capacity of 2000 UEs per carrier. For the LTE RAT a capacity of 15 Mbps per UE is assumed, corresponding to a baseline capacity of 2000 UEs per carrier.

A first example assumes an M2M scenario, for example involving sensor UEs transmitting measured data, e.g., information about temperature. Further, an even distribution of UEs in all three sectors of the three ANs, is assumed. Each UE is assumed to have a traffic profile with CP traffic of 0.5 kbps in the UL traffic direction and 0.5 kbps in the DL traffic direction and UP traffic pf 5 kbps in the UL traffic direction and 4 kbps in the DL traffic direction. Further, it is assumed that the number of UEs per sector is 1000, corresponding to a total population of 9000 UEs in the three sectors of the three ANs.

400 When now assuming that before application of the CUPS optimization algorithm, the UEs send CP traffic and UP traffic through the LTE RAT, using one TX port and one RX port, the consumption per sector is: 10000 mW+10000 mW+[1000*(4 kbps+5 kbps+0.5 kbps+0.5 kbps)]*10 mW/kbps=120000 mW=120 W. When further the GPRS carriers also active, there is an additional of contribution 20 W per sector, giving a consumption of 140 W per sector. Accordingly, the total consumption of all three ANs would be 3*3*140 W=1.26 MW.

400 When now assuming that the CUPS optimization algorithmassigns the GPRS carriers to the role of conveying the CP traffic while the LTE carriers are used only for the UP traffic, it becomes possible to used only one three-sector AN for maintaining coverage for the CP traffic, because the GPRS RAT has a wider range. Accordingly, the GPRS carriers of the other ANs can be turned off. The consumption for CP traffic thus amounts to 9000*(0.5 kbps+0.5 kbps)*2 mW/kpbs=18000 mW=18 W. The consumption for LTE UP traffic per sector is 1000*(4 kbps+5 kbps)*10 mW/kbps=90000 mW=90 W. The consumption for the total UP traffic for all sectors of all three ANs thus is 3*3*90 W=810 W. The total consumption for CP traffic and UP traffic thus is 810 W+18 W=828 W. As compared to the situation before optimization, there is thus a power saving of 432 W.

A second example assumes a mobile broadband scenario, with the same setup of three ANs, however with a different traffic profile of the UEs. In the second example, each UE is assumed to have CP traffic of 1 kbps in the UL traffic direction and 1 kbps in the DL traffic direction, and UP traffic of 100 kbps in the UL traffic direction and 400 kbps in the DL traffic direction. The number of UEs per sector is assumed to be 25, however in an uneven distribution with one of the sectors being unoccupied.

When all UEs operate on LTE carriers, the consumption can be calculated as follows: A total baseline consumption for operating all RF ports in all RUs of three sectors of three ANs is 3*3*40=360 W. The total population of UEs is 25*2*3=150. The additional cost for transmission and reception of CP traffic and UP traffic of the UEs is 150*(100 kbps+400 kbps+1 kbps+1 kbps)*10 mW/kbps=753000 mW=753 W. The total consumption is therefore 360 W+753 W=1113 W=1.113 MW.

400 When now assuming that the CUPS optimization algorithmdecides to turn of the RF ports of the unpopulated LTE sectors and to activate the GPRS carriers of one AN covering the entire UE population for CP traffic, the consumption is as follows: A total baseline consumption for operating the active RF ports is 3*2*40 W+40 W*3=360 W. The additional cost for transmission and reception of CP traffic and UP traffic of the UEs is 150*[(100 kbps+400 kbps)*10 mW/kbps]+150*[(1 kbps+1 kbps)*2 mW/kbps]=750000 mW+600 mW=750600 W=750.6 W. The total consumption is therefore 360 W+750.6 W=110.6 W=1.106 MW. As compared to the situation before optimization, there is thus a power saving of 2.4 W. An alternative optimization result could be that the GPRS carriers are not activated, and only unused LTE carriers are inactivated. The consumption is then as follows: A total baseline consumption for operating the active RF ports is 3*2*40 W=240 W. The additional cost for transmission and reception of CP traffic and UP traffic of the UEs is 150*(100 kbps+400 kbps+1 kbps+1 kbps)*10 mW/kbps=753000 mW=753 W. The total consumption is therefore 240 W+753 W=993 W. As compared to the situation before optimization, there is thus a power saving of 120 W.

400 As illustrated by the above examples, by considering the predicted traffic profiles of UEs and the population of carriers resulting from the predicted mobility, the CUPS optimization algorithmcan achieve significant power saving.

6 FIG. 6 FIG. 200 100 270 230 shows a flowchart for illustrating a method of controlling transmissions in a wireless communication network. The method ofmay be utilized for implementing the illustrated concepts in a node of the wireless communication network, e.g., in a node implementing the above-mentioned RAF. As mentioned above, such node could be implemented by one or more access nodes, such as the above-mentioned access nodes, or by a management node, such as the above-mentioned management node, or a control node, such as the above-mentioned control node.

6 FIG. 6 FIG. Accordingly, at least some steps of the method ofmay be performed by one or more nodes of a RAN of the wireless communication network. Alternatively or in addition, at least some steps of the method ofmay be performed by one or more nodes of a CN of the wireless communication network. Further, it would also be possible that at least some of the steps of the method are performed by one or more nodes outside the RAN or CN of the wireless communication network, e.g., in one or more nodes hosted by a third-party cloud provider.

6 FIG. 6 FIG. 6 FIG. If a processor-based implementation of such node performing at least some steps of the method ofis used, at least some of the steps of the method ofmay be performed and/or controlled by one or more processors of the node. Such node may also include a memory storing program code for implementing at least some of the below described functionalities or steps of the method of.

610 410 100 At step, throughput data representing data throughput of wireless communication devices may be obtained, such as the above-mentioned throughput data provided as input to the first ML model. The throughput data can be collected from access nodes of the wireless communication network, e.g., from the above-mentioned access nodes.

620 610 410 620 At step, a first ML model for prediction of data throughput of wireless communication devices may be trained, e.g., using at least a part of the throughput obtained at step. The above-mentioned first ML modelis an example of the first ML model trained at step. The first ML model may be based on a neural network, e.g., an RNN.

630 420 At step, mobility data representing mobility of wireless communication devices may be obtained, such as the above-mentioned mobility data provided as input to the second ML model. The mobility data can be collected from one or more mobility management nodes of the wireless communication network, e.g., from an MME or AMF.

640 610 420 630 At step, a second ML model for prediction of mobility of wireless communication devices may be trained, e.g., using at least a part of the throughput obtained at step. The above-mentioned second ML modelis an example of the second ML model trained at step. The second ML model may be based on a neural network, e.g., an RNN.

650 610 620 610 At step, data throughput is predicted. In particular, for a set of access nodes of the wireless communication network, data throughput is predicted for UP traffic and CP traffic generated by wireless communication devices which are served by the access nodes on a plurality of wireless carrier signals. The data throughputs may for example be predicted based on at least a part of the throughput data obtained at step. In particular, the data throughputs may be predicted by the first ML model trained at step, and this prediction by the first ML model may be accomplished based on at least a part of the throughput data obtained at step.

650 The prediction of the data throughput at stepmay thus be based on a first ML model trained based on of first data representing data throughputs observed with respect to the set of access nodes.

660 630 640 630 At step, mobility of the wireless communication devices is predicted. The mobility may for example be predicted based on at least a part of the mobility data obtained at step. In particular, the mobility may be predicted by the second ML model trained at step, and this prediction by the second ML model may be accomplished based on at least a part of the mobility data obtained at step.

660 The prediction of the mobility at stepmay thus be based on a second ML model trained based on second data representing mobility observed with respect to the set of access nodes. The mobility may be observed in connection with attachment of a wireless communication device to an access node of the set, detachment of a wireless communication device to an access node of the set, handover of a wireless communication device from an access node of the set, and/or handover of a wireless communication device to an access node of the set.

670 650 660 320 At step, based on the data throughput predicted at stepand the mobility predicted at step, it is decided for at least some of the plurality of wireless carrier signals whether the wireless carrier signal is assigned either to serving UP traffic, to serving CP traffic, or to serving both UP traffic and CP traffic. Accordingly, the roles of at least some of the wireless carrier signals may be reassigned among the roles of CP carrier, UP carrier, and CP/UP carrier. Further, based on the predicted data throughput and the predicted mobility, it can be decided for at least some of the plurality of wireless carrier signals whether to turn off transmission on the wireless carrier signal. Such turning off of transmission on the wireless carrier signal may involve turning off a port of the access node which transmits or receives the wireless carrier signal, e.g., like one or more of the above-mentioned RF ports. Such decision may also consider possible reassignments of the wireless carrier signals as CP carrier, UP carrier, or CP/UP carrier.

670 The decision of stepmay be based on an assignment optimization algorithm aiming at minimization of energy consumption by the set of access nodes, such as the above-mentioned CP-UP assignment optimization stage. Various kinds of optimization algorithm could be used, e.g., an optimization algorithm based on an ML model. Such ML model could be trained using RL, based on actions corresponding to role assignments of the wireless carrier signals and rewards based on changes in energy consumption reported by the access nodes. Accordingly, the assignment optimization algorithm could be trained based on data representing energy consumption reported with respect to the set of access nodes.

The wireless carrier signals are allocated to different RATs. Such different RATs may for example include the NR RAT, the LTE RAT, the UMTS RAT, and/or the GPRS RAT. Further, the wireless carrier signals may be allocated to different transmission directions. For example, a first subset of one or more of the wireless carrier signals can each be allocated to a UL transmission direction. Further, wherein a second subset of one or more of the wireless carrier signals can each be allocated to a DL transmission direction. Further, a third subset of one or more of the wireless carrier signals can each be allocated to both the DL and UL transmission directions.

7 FIG. 6 FIG. 700 700 100 270 230 700 200 700 710 610 700 720 620 700 730 630 700 740 640 700 750 650 700 760 660 700 770 670 shows a block diagram for illustrating functionalities of a network nodewhich operates according to the method of. The network nodemay for example implement a RAN node, such as the above-mentioned access nodes, or a CN node, such as the above-mentioned management systemor control node. The nodemay correspond to the above-mentioned RAF. As illustrated, the network nodemay be provided with a moduleconfigured to obtain throughput data, such as explained in connection with step. Further, the network nodemay be provided with a moduleconfigured to train an ML model for predicting throughput, such as explained in connection with step. Further, the network nodemay be provided with a moduleconfigured to obtain mobility data, such as explained in connection with step. Further, the network nodemay be provided with a moduleconfigured to train an ML model for predicting mobility, such as explained in connection with step. Further, the network nodemay be provided with a moduleconfigured to predict throughput, such as explained in connection with step. Further, the network nodemay be provided with a moduleconfigured to predict mobility, such as explained in connection with step. Further, the network nodemay be provided with a moduleconfigured to decide on assignment of wireless carrier signals, such as explained in connection with step.

700 700 700 It is noted that the network nodemay include further modules for implementing other functionalities, such as known functionalities of a management system. Further, it is noted that the modules of the network nodedo not necessarily represent a hardware structure of the network node, but may also correspond to functional elements, e.g., implemented by hardware, software, or a combination thereof.

8 FIG. 8 FIG. 800 800 100 270 230 800 200 200 800 illustrates a processor-based implementation of a network nodewhich may be used for implementing the above-described concepts. The network nodemay for example implement a RAN node, such as the above-mentioned access nodes, or a CN node, such as the above-mentioned management systemor control node. The nodemay correspond to the above-mentioned RAFor implement at least a part of the functionalities of the RAF. In some scenarios, also a system of multiple network nodeswith structures as illustrated inmay be used implementing the above-described concepts, e.g., in a cloud-architecture where such multiple network nodes cooperate.

800 810 810 As illustrated, the network nodeincludes one or more interfaces. These interfacesmay for example be used for enabling communication with one or more other network nodes.

800 850 810 860 850 810 850 860 800 860 Further, the network nodemay include one or more processorscoupled to the interface(s)and a memorycoupled to the processor(s). By way of example, the interface(s), the processor(s), and the memorycould be coupled by one or more internal bus systems of the network node. The memorymay include a Read-Only-Memory (ROM), e.g., a flash ROM, a Random Access Memory (RAM), e.g., a Dynamic RAM

860 870 880 860 850 6 FIG. (DRAM) or Static RAM (SRAM), a mass storage, e.g., a hard disk or solid state disk, or the like. As illustrated, the memorymay include softwareand/or firmware. The memorymay include suitably configured program code to be executed by the processor(s)so as to implement the above-described functionalities of a network node, such as explained in connection with.

8 FIG. 800 860 800 860 It is to be understood that the structures as illustrated inare merely schematic and that the network nodemay actually include further components which, for the sake of clarity, have not been illustrated, e.g., further interfaces or processors. Also, it is to be understood that the memorymay include further program code for implementing known functionalities of a network node, e.g., known functionalities of a management system for a 3GPP network. According to some embodiments, also a computer program may be provided for implementing functionalities of the network node, e.g., in the form of a physical medium storing the program code and/or other data to be stored in the memoryor by making the program code available for download or by streaming.

As can be seen, the illustrated concepts may be used for improving power efficiency of CUPS. In particular, by considering predicted data throughput and predicted mobility, role assignments of carriers as CP carrier, UP carrier, or CP/UP carrier may be efficiently optimized to minimize overall energy consumption of the considered access nodes.

It is to be understood that the examples and embodiments as explained above are merely illustrative and susceptible to various modifications. For example, the illustrated concepts may be applied in connection with various wireless communication network technologies, without limitation to the above-mentioned examples. Further, the illustrated CUPS optimization algorithm may be used for access node populations of various sizes, e.g., access nodes of a tracking area or even all access nodes of a certain operator.

Moreover, it is to be understood that the above concepts may be implemented by using correspondingly designed software to be executed by one or more processors of an existing device or apparatus, or by using dedicated device hardware. Further, it should be noted that the illustrated nodes may each be implemented as a single device or as a system of multiple interacting devices or modules, e.g., as a cloud system.